Incidence of lower extremity amputations in the diabetic compared with the non-diabetic population: A systematic review

Lower extremity amputation (LEA) in patients with diabetes results in high mortality, reduced quality of life, and increased medical costs. Exact data on incidences of LEA in diabetic and non-diabetic patients are important for improvements in preventative diabetic foot care, avoidance of fatal outcomes, as well as a solid basis for health policy and the economy. However, published data are conflicting, underlining the necessity for the present systematic review of population-based studies on incidence, relative risks and changes of amputation rates over time. It was conducted according to the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines. Nineteen out of 1582 studies retrieved were included in the analysis. The incidence of LEA in the diabetic population ranged from 78 to 704 per 100,000 person-years and the relative risks between diabetic and non-diabetic patients varied between 7.4 and 41.3. Study designs, statistical methods, definitions of major and minor amputations, as well as the methods to identify patients with diabetes differed greatly, explaining in part these considerable differences. Some studies found a decrease in incidence of LEA as well as relative risks over time. This obvious lack of evidence should be overcome by new studies using a standardized design with comparable methods and definitions. Systematic review registration number PROSPERO CRD4201501780


Introduction
The global prevalence of diabetes mellitus (DM) has risen to 8.8% in 2015, which corresponds to 415 million patients [1].This leads to increasing numbers of individuals with diabetic foot disease, up to 75% of lower extremity amputations (LEAs) being performed in these patients [2,3].LEAs reduce the quality of life [4] and increase mortality [5][6][7] as well as medical costs [8].Early initiatives [9][10][11] persued the goal to reduce the number of LEAs in diabetic patients.However, ensuing epidemiological studies showed marked variations in the incidence, relative risks and time trends of LEA compared with the non-diabetic population, owing to differences in study design and methodological approaches [3,[12][13][14].Furthermore, there is no generally accepted definition of major or minor amputation [12,15,16] and the unambiguous identification of a diabetic person is conducted in a large variety of ways [16][17][18][19].Also, incidence data differ largely depending on whether they are based on the number of amputations, hospitalizations or amputees.For example, the incidence rates per 10,000 diabetic patients per year were reported with 158 amputations, 101 hospitalizations and 87 amputees [20].Finally, statistical methods differ largely between the studies, as they analysed crude or adjusted incidence rates.[13,18,21].Reliable incidence rates of LEA in diabetic and non-diabetic populations are of utmost global importance for further improvements in the care of diabetic patients, the avoidance of fatal outcomes and for decisions relating to health policy and economy.Some reviews on this topic have been published [22][23][24][25][26][27]; however, they had some limitations with respect to definitions of LEA, at-risk population and statistical methods.Therefore, we felt that a systematic review was overdue.The aims of this systematic review were (1) to analyse the incidence of LEA in the population with and without diabetes as well as the corresponding relative risks, and (2) to investigate time trends.

Methods
The systematic review was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [28] (S1 Table ).A study protocol with the registration number PROSPERO CRD42015017809 (S1 Text) has been published [29].

Search strategies and selection criteria
Literature was searched systematically in the international biomedical literature databases MEDLINE, EMBASE, Web of Knowledge and publisher databases Journals@OVID and Scien-ceDirect until December 2014 with no retroactive time limit.Database-specific controlled terms (MeSH, EMTREE) and additional free-text terms were used.The search terms (combined by Boolean operators) were amputation, amputee (search component "intervention"), lower extremity, foot, feet, limb, etc. (search component "problem") and epidemiology, prevalence, incidence, frequency, population survey, survey data, administrative data, community data etc.(search component "epidemiologic studies").The systematic search is based on a linear block-building model.Some cherry-picking strategies were added.Potentially eligible studies in reference lists of review articles and relevant studies were identified by additional handsearch.The detailed search strategies are provided as supporting information (S2 Text).
Full-text articles were included if they met inclusion criteria regarding types of studies, types of population and the main outcome regardless of the type of amputations, age, sex and ethnicity.
Types of studies.All population-based studies analysing incidence rates in the diabetic versus the non-diabetic population with reported relative risks were included.
Types of population.(1) The population at risk had to be defined by official statistics, which means nationwide data or for example all inhabitants of a defined region or all insurants of a statutory health insurance.(2) The diabetic population had to be precisely described (e.g.register, estimation based on age-sex-specific prevalence data).Individuals without diabetes were considered with the aim of comparing incidences between the diabetic and non-diabetic populations.
Outcomes (definition of LEA, epidemiological measures).The main outcome was analysed depending on the level of amputation: I. any LEA (both major and minor amputations); II.major amputations; III.minor amputations.This was always done in terms of reporting the incidence of LEA: a) person level: only one amputation per person (first or highest); b) case level: data based on hospital discharge rates.This may be several hospitalizations per person in the same calendar year; nevertheless, as a rule only one amputation per person per admission is considered; c) procedure level: all amputations per person are taken into account.
The epidemiological measure of the main outcome was the incidence rate (IR) of LEA among patients with diabetes and among persons without diabetes.In order to compare incidence rates of LEA between the diabetic and non-diabetic population, the relative risks (RRs) were taken into account.Furthermore the attributable risk (AR) and the population attributable risk (PAR) were considered where available.AR is the proportion of LEA among persons with diabetes that is attributable to diabetes.PAR is the proportion of LEA in the whole population that is attributable to diabetes.
Studies were excluded if: (a) they solely reported incidences of LEA among persons with diabetes without comparison to non-diabetic persons; (b) incidence rates were reported in relation to the total (diabetic and non-diabetic) population and not exclusively using the diabetic population as a population at risk; (c) exclusively crude incidence rates were reported; (d) studies were published in a language other than English.

Data collection and extraction
Four authors (H.C., A.I., T.K. and B.S.) independently screened all the retrieved titles and abstracts to identify potentially eligible articles.Full-text screening was performed by four authors (M.N., H.C., T.K. and B.S.).Disagreements were resolved by discussion.Data extraction was performed by M.N., H.C. and B.S., including information about first author, publication year, country, study period, study design, study populations (diabetic and non-diabetic), definition of LEA and study results.The reported IR was recalculated as IR per 100,000 PY if not originally reported as such.

Quality assessment
The quality of the eligible studies was assessed by two independent reviewers (M.N. and B.S.) in consideration of study limitations, risk of bias and the degree of imprecision (missing information according to absolute number of cases or missing confidence interval) using the modified checklist (S2 Table) adapted to Methodological Evaluation of Observational Research (MORE) [30], the Scottish Intercollegiate Guidelines Network (SIGN) [31], and the Cochrane approach Study Quality Guide [32].This tool was used to define criteria based on clinical and epidemiological expertise and studies with quality ranked as high, acceptable or low according to the recommendations of SIGN [31].Detailed information can be found in the study protocol [29].Only studies with high or acceptable quality were included in the review.

Statistical methods
The results (IR, RR, AR, and PAR) from included studies were presented as age-sex-adjusted estimates.All estimates were presented with 95% confidence intervals (95% CI) if available.In either case, the number of events of LEA was included in the results tables.We described time trend as "descriptive" if the reviewed studies only reported annual incidence of LEA without using appropriate multivariate regression models.Due to the high heterogeneity of the included studies no meta-analysis was performed.

Results
1582 citations were initially retrieved, from which 19 papers were included in this review.The selection procedure is presented in Fig 1.

Quality assessment
Using the "Methodology Checklist critical appraisal" tool we considered 10 studies to be of high quality, nine to be of acceptable quality, and four studies to be of low quality.The latter were excluded (Fig 1).The important critical points were: the unclear definition of LEA, the missing data concerning confidence intervals, and the absolute number of LEAs.
Ethnic differences: Gujral et al. reported lower IRs among patients of Asian ethnicity compared with white Caucasians [40].
Relative, attributable and population attributable risks between the diabetic and non-diabetic population: The RRs ranged from 7.4 (95% CI 6.3-8.7)[12] to 26 (95% CI 17-39) [2] (Table 2, Fig 3).The RRs in diabetic compared with non-diabetic persons decreased with increasing age [12,39].Two studies reported higher RRs among men [12,39] and one among women [37].Based on the studies available it was not possible to estimate the ethnic differences regarding the RRs.
Time trends of the incidence rates: Out of three publications that analysed the time trend [2,13,16], only one study [16] was able to find a significant reduction in IRs of LEAs in the diabetic population: RR per calendar year 0.976 (95% CI 0.958-0.996),whereas in the non-diabetic population the secular trend was unchanged (Table 2, Fig 2).
Time trends of the relative risks: The study by Lombardo et al. [13] found no change in RRs whereas the study by Trautner et al. [16] found a significant reduction of RRs during the study period.
Two studies [3,36] reported the reduction of RRs between diabetic and non-diabetic populations with increasing age.Concerning gender differences, the study by Almaraz et al. [3] described the higher RRs between 1998-2003 among women and between 2004-2006 among men, whereas the study by Van Houtum et al. found no differences [35].Interestingly, one US study which presented data stratified by ethnic origin found lower RRs among African-American persons than among non-Hispanic whites and Hispanic persons [11].One study from the 1990s reported higher RRs in the USA than in Europe [34].In contrast, since the 2000s the reported RRs were higher in Europe [3,33] than in the USA [19].
Time trends of the incidence rates: Three studies described the time trend in the diabetic population with contradictory results [3,19,33].The study by Gregg et al. [19] reported a significant reduction of IRs between 1990-2010 in the USA.In contrast, the study by Almaraz et al. [3] reported a significant increase of IRs between 1998-2006 in Andalusia, Spain, while a study by Buckley et al. [33] showed non-significant growth of IRs in Ireland between 2005-2009.All studies found decreased IRs in the non-diabetic population over time.Time trends of the relative risks: the study by Gregg et al. [19] reported a substantial decrease of RRs over time, while the study by Almaraz [3] described the increasing RRs during the study period and the study by Buckley et al. found no significant changes [33].
I.c.Incidence of any LEA: All LEAs per person.Only one study in the UK described IRs regarding all LEAs per person [38].This study reported a reduction of both IRs in the diabetic population from 564.3 to 176.0 per 100,000 PY as well as RRs from 46 (95% CI 25.7-90.6)to 7.7 (95% CI 4.99-12.9)between 1995-1996 and 1999-2000 [38] (Table 2).
Time trends of the incidence rates: Among the diabetic population only those studies which estimated one major amputation per person found a significant reduction of IRs of major LEAs during the study period (S1 Fig) [13,16,17,38].Among the non-diabetic population some studies also reported decreased secular trends [3,13,17].
Time trends of the relative risks: Among the studies that counted one major LEA per person and analysed a time trend, two found significantly decreased RRs over time [16,17] whereas one found no change in RRs [13].Another study counting all major LEAs also showed a descriptive reduction during a 5 year period from 35.5 (95% CI 18.9-76.8)to 5.0 (95% CI 2.82-9.43)respectively [38].
III. Incidence of minor LEA.Three studies estimated IRs of minor LEA counting one LEA per person [13,15,39], three studies were based on hospitalization rates [3,13,33], and two described all minor LEAs per person [15,38].
Incidence rates: As expected, the highest IRs of minor LEA in the diabetic population were found in the group counting all LEAs per person with a maximum 362.9 (95% CI NA) per 100,000 PY in the UK, 1998-1999 [38], and the lowest IRs of 85.5 (95% CI NA) per 100,000 PY in Italy, 2008 [13] were described in the group counting one LEA per person.In the non-diabetic population IRs were not higher than 10 per 100,000 PY and were reasonably comparable [13,15,33].Two studies found higher IRs of minor LEAs among men than women [15,39].Relative risks between diabetic and non-diabetic populations: RRs between diabetic and non-diabetic populations were generally higher than those for major LEAs and ranged between 15.72 (95% CI NA) in 1993-1994 in UK [39] and 40.9 (95% CI 33.0-50.7) in 2009 in Ireland [33].
Time trends of the incidence rates: Among persons with diabetes one study found no significant change of IRs of minor LEA over time [13], while others showed a significant [3] and a non-significant increase [33] of IRs.Among persons without diabetes one study described increased [13] and two studies unchanged secular trends [3,33].
Time trends of the relative risks: Only the study by Lombardo et al. analysed a time trend of RRs and found no significant changes during the study period [13].

Discussion
This systematic review, comprising 19 studies, demonstrates considerable variations in incidence of LEA as well as in relative risks of amputations in the diabetic compared with the nondiabetic population.This variation may probably be due to the large heterogeneity of the studies included.

Main findings
In 1989, the St. Vincent Declaration set the goal to reduce the incidence of LEA by half within five years [9] and thus to approach the incidence in the non-diabetic population.Still there remains uncertainty as to whether this goal has been achieved.The time trend of both incidence rates and relative risks in the diabetic as well in the non-diabetic population varied with different definitions of LEA (any, minor, major) and methods of recording (one LEA per person or more).The previous reviews [25,27] already highlighted the importance to take into account the methodological discrepancies between studies by interpreting the published data.
Incidence rates of LEA.The studies reviewed showed considerable variation in incidence rates of LEA in both the diabetic and non-diabetic populations.Nevertheless, there were some patterns that can be described.
Gender differences: Some studies found higher IRs among men than women in both the diabetic and the non-diabetic population.The gender-relative risks ranged from 1.5 [37] to 3.0 [3] for any LEAs and from 1.56 [39] to 5.0 [15] for major LEAs.Moreover, men were younger at the time of LEA then women [2,3,12,16,17,39].This strong association between risk of LEAs and male gender was described in earlier publications [47].The increased IRs among men may be explained by environmental factors such as smoking as well as by higher prevalence of peripheral vascular disease, peripheral neuropathy and diabetic foot ulceration [35,47], but not by healthcare factors [47].
Ethnic differences: Two studies, both from the USA, found higher IRs among African-Americans than white Caucasians in both the diabetic and non-diabetic populations [11,19].Risk factors such as smoking, low socio-economic status, and poor access to healthcare may contribute to observed ethnic disparities [27,48].Moreover, the African-American ethnicity per se could be an independent risk factor for LEA [27,49].Nevertheless, it is probably true only for the African-American ethnicity in the USA.One recent study by Holman found no significant differences in the crude incidence rates of LEA between black and white residents in England [21].The reason for these contrasting findings could be due to the organization of the healthcare systems: private in the USA and a public healthcare system in the UK.Interestingly, that one reviewed study demonstrated a lower incidence of LEA in Asian diabetic patients than in white patients [40].This finding is in line with the study by Holman mentioned above [21].The lower prevalence of PAD [50] or neuropathy [51] among Asians may be possible protective factors.International and regional differences: An international comparison showed a higher incidence of LEA in both diabetic and non-diabetic populations in the USA than in Europe until the early 2000s [19,34]; in later years the incidences were converging [19].Regional disparities within a country were found in both the USA [41] and Europe [35], being more pronounced in the diabetic than in the non-diabetic population [35,41].The RRs between diabetic and non-diabetic populations varied widely across regions, too [35,41].Differences in the regional distribution of Asian or Black ethnicity [21,52], differences in access to healthcare [41,52] as well as in the organization of diabetes care and the quality of foot-care centers [21,41,52] were described as potential influencing factors.
Time trends: Only studies that analysed the incidence rates of one major LEA per person found a significant decrease in the incidence of LEAs in the diabetic population over time (Table 2).No clear trends in the incidence of LEA could be observed in studies that analysed more than one major LEA or in studies that analysed minor LEA irrespective of the counting methods (Table 2).For studies among the non-diabetic population no time trend could be observed (Table 2).We consider it important to analyse the time trend of LEAs depending on the extent of amputations, as the clinical distinction and objectives of performing major and minor LEAs are different.The reduction regarding incidence of major LEA in the diabetic population could be explained in particular by better organised multidisciplinary care for patients with diabetic foot [13,16,38] but also by improvements in diabetes care [13,15,19], tighter control of hypertension and hypercholesterolemia [15], and reduced tobacco and alcohol consumption among persons with diabetes [15].Improvements in bypass surgery as well as new endovascular revascularization techniques are important factors for the reduction of LEA incidence among patients with peripheral vascular disease [53][54][55][56].However, not all reviewed studies showed significant reduction of LEA incidence in the non-diabetic population.The reduced incidence of major LEA can therefore be explained only in part by the success of vascular surgery [57].
Relative risks between the diabetic and non-diabetic populations.In general, RRs between diabetic and non-diabetic persons for minor LEAs were more than twice as high as for major LEAs [13,33].Similarly, RRs based on hospital discharge rates were higher than in studies counting one LEA per person [13].Some studies reviewed reported decreased RRs with increased age of patients at the time of LEA [3,12,36,39].This could be explained by the substantial growth in IRs of LEAs due to other reasons among older non-diabetic patients when compared with diabetic patients from the same age group.Regarding gender differences, the results were not fully consistent: most studies reported higher RRs among men [3,12,17,39], one among women [3,37] and another described similar RRs [35].The RRs among African-Americans were lower than among non-Hispanic whites and Hispanics [11].Until the 2000s the RRs were higher in the USA than in Europe [34], although that trend seems now to have reversed [3,19,33].Most studies that analysed one major LEA showed that the relative risks among persons with diabetes has been decreasing over time compared with persons without diabetes [16,17,38].The results on time trends of RRs of LEA in the other groups of studies were contradictory (Table 2).

Risk of bias within studies
Selection bias regarding the study population was minimized through the restriction to population-based studies.At the same time, some sources for information bias were detected.
Firstly, it appears reasonable to assume that the registration of major LEAs in hospitals is complete (owing to the fact that hospital reimbursement often depends on the major amputation procedures).However, the number of minor LEAs based on the hospital data could be underestimated due to incomplete documentation, since documenting more than one minor LEA during one hospital stay does not normally trigger higher refunds and may therefore be neglected.Also, autoamputations or minor LEAs performed outside the operating theatre are presumably underreported.
Secondly, in most studies the definition of diabetes among amputees was based on the coding of diabetes diagnosis in admission or discharge records.However, it was shown that diabetes diagnosis based on hospital admissions data can lead to underreporting of diabetes by up to 15% [13,18,58] and thus to a considerable underestimation of the incidence of LEAs in the diabetic population.
Thirdly, most studies used survey data for the estimation of a "diabetic population at risk" or created a special algorithm to identify persons with diabetes.However, these methods could lead to some misclassification, in particular underestimation when a diabetes diagnosis is not documented [12,16].Furthermore, patients with undiagnosed diabetes might have been misclassified as non-diabetic patients.Only in one study by Calle-Pascual were patients with undiagnosed diabetes also taken into account [15].Most studies also used constant diabetes prevalence during the study period.In this case, instant diabetes prevalence is underestimated [2,3,16].Moreover, the change in 1997 in the diagnostic criteria for diabetes from 140 mg/dl (7.8 mmol/l) to 126 mg/dl (7.0 mmol/l) fasting plasma glucose [59] led to an increase of the diabetic population due to the inclusion of less severe stages of the disease, and this must also be taken into consideration when interpreting the results.
Fourthly, the vast majority of the studies reviewed were not able to distinguish between type 1 and type 2 diabetes.It is therefore not clear if the incidence and the time trend of LEAs among patients with type 1 diabetes differ from those among patients with type 2 diabetes.
Finally, low absolute numbers of LEAs in small study populations may cause strong variations in the incidence rates [2,16].This complicates the appropriate interpretation of the data.

Risk of bias across studies
Due to the fact that only articles published in the English language were reviewed, publication (language) bias could not be ruled out.Although we searched five databases, we cannot guarantee that some related papers may not have been identified.However, we did check the reference lists of reviewed articles to identify relevant studies.Furthermore, differences in methodological approaches across studies can lead to limited comparability.The studies reviewed used different definitions of major amputations: through or proximal to the ankle joint [39], any LEA above the ankle joint [2,13,16,41]; any LEA proximal to the midtarsal level (International Consensus of the Diabetic Foot) [12]; through or proximal to the tarsometatarsal joint (GLEAS protocol) [15,38].However, hindfoot amputations such as Syme's, Chopart or Lisfranc may constitute a relevant proportion of all amputations [60].For the comparability of future studies it is therefore clearly important to find an unequivocal differentiation between major and minor LEA, since the percentage of amputations "in between" is not marginal.
Differences were also observed in the way the studies reviewed reported the causes of LEAs.Most studies analysed non-traumatic LEAs, three estimated incidence of all LEAs regardless of their cause [12,17,18], a number analysed "diabetes-related" (not traumatic and not tumourrelated) LEAs [3,13,39,40] and one study from Finland analysed only LEAs attributable to peripheral vascular disease [37].However, it was shown that traumatic LEAs are less frequent in diabetic than in non-diabetic populations (6.5% vs. 15%) [18].
With regard to the divergent definition of diabetic population at risk, only few studies used data based on actual registers, which means almost complete data collection [17,37].In contrast, most studies used the estimated prevalence of diabetes.Moreover, most studies reviewed the estimated diabetic population at risk as a percentage of patients with known diabetes, one study as a percentage of patients with known and unknown diabetes [15], and two used the denominator based on the number of patients with treated diabetes [17,18].Due to the high prevalence of unknown type 2 diabetes [1, 61,62], it is expected that the incidence of LEAs with the denominator based on known and unknown diabetes will be considerably lower than where the denominator is based on a population with known diabetes.Where the denominator is based on drug-treated diabetes it should be taken into account that the percentage of diet-treated type 2 diabetes patients could amount to up to 25% [63].
Finally, specific characteristics of the study population such as age or ethnicity could also influence the results.Most studies had no age restriction, but some studies analysed the population aged !20 years [33] or aged !30 years [3,19], and the study by Wrobel et al. [41] analysed the Medicare population aged !65 years.Study populations from the USA [11,19,34,41] are clearly distinguished from European studies by a high proportion of African-American and Hispanic persons.

Strengths and limitations
The selection of studies for this systematic review was based on a systematic search approach with clearly determined search strategies.Two independent reviewers screened the articles and performed the data extraction.We included only those studies reporting IRs of LEA within the population at risk, i.e. the diabetic population.The advantage of this method over IRs of LEA within the general population is that the results are not influenced by changes in the prevalence of diabetes.Moreover, we analysed incidences of LEA in the diabetic population in separate groups according to definition of LEA as well as the method of recording and study design (study characteristics).This approach allows comparison of the studies despite a high degree of heterogeneity.Nevertheless our review has some limitations.Although seven databases were searched, we cannot rule out having missed any relevant studies due to publication bias.Furthermore, studies that were published in languages other than English were excluded.Most studies reporting on IRs of LEA among patients with diabetes within the diabetic population were conducted in economically developed areas such as the USA and Europe, and thus do not represent a worldwide perspective.

Conclusion
This comprehensive review demonstrates the considerable variation in incidence of LEA among the diabetic population, probably partly due to a large heterogeneity of identified studies.As expected, the incidence of LEA was higher in the diabetic than in the non-diabetic population.Most studies found a higher incidence of LEA among male diabetic patients.Black and Hispanic patients have a higher risk of LEA than white individuals, Asian patients, however, do not.Studies that analysed one major LEA found decreased incidence rates among the diabetic population as well as corresponding relative risks over time.Among studies with different study design, the current data on time trends for incidence rates as well as relative risks between diabetic and non-diabetic populations are contradictory.The studies reviewed showed high regional and international differences with respect to both incidence and relative risks of LEA.A comparison was difficult due to the lack of consensus between the studies' methods.We recommend that future studies analysing the incidence and relative risks of LEA in the diabetic population should use a comparable study design regarding anatomic definition, cause and recording of LEAs as well as the population at risk.

Fig 1 .
Fig 1. Flowchart of the systematic review process.* The quality of the eligible studies was assessed using the modified checklist (s.S2 Table).

Study period, design and population Diabetic population at risk, number at risk (n) Data sources for diabetic prevalence Definition of LEA by anatomical level Counting of LEA Cause for LEA Data sources for LEA Time trend
[46]R diabetic population at risk; LEA lower extremity amputation; AR attributable risk; PAR population attributable risk; DD diabetes duration; DM diabetes mellitus; ITDM insulin-treated diabetes mellitus; NITDM non-insulin-treated diabetes mellitus; PVD peripheral vascular disease; GLEAS Global Lower Extremity Amputation Study[46].https://doi.org/10.1371/journal.pone.0182081.t001